Do Bankruptcy Protection Levels Affect Households' Demand for Stocks?
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Banco de México
Documentos de Investigación
Banco de México
Working Papers
N° 2021-03
Do Bankruptcy Protection Levels Affect Households'
Demand for Stocks?
Mariela Dal Borgo
Banco de México
May 2021
La serie de Documentos de Investigación del Banco de México divulga resultados preliminares de
trabajos de investigación económica realizados en el Banco de México con la finalidad de propiciar el
intercambio y debate de ideas. El contenido de los Documentos de Investigación, así como las
conclusiones que de ellos se derivan, son responsabilidad exclusiva de los autores y no reflejan
necesariamente las del Banco de México.
The Working Papers series of Banco de México disseminates preliminary results of economic
research conducted at Banco de México in order to promote the exchange and debate of ideas. The
views and conclusions presented in the Working Papers are exclusively the responsibility of the authors
and do not necessarily reflect those of Banco de México.Documento de Investigación Working Paper
2021-03 2021-03
Do B a nkrupt c y P r o t e c t i o n L e v e l s Af f e c t Ho u s e h o l d s'
De m a n d f o r St o c k s ? *
Mariela Dal Borgo†
Banco de México
Abstract: This paper examines empirically the effect of the level of personal bankruptcy protection in
the US on households' demand for financial assets. A Chapter 7 bankruptcy allows protecting the home
equity up to a certain limit or "exemption". Previous literature shows that such exemption biases
investment towards home equity. This paper tests whether it also lowers investment in stocks, which are
not protected in bankruptcy. Using an instrumental variable approach, I estimate a lower stock market
participation when the home equity is below the exemption, but the result is not robust, and households
at higher risk of bankruptcy do not exhibit a stronger response. Moreover, investment in home equity is
not higher when the home is fully protected. These findings suggest no substantial portfolio distortions
from the level of home equity that is protected in bankruptcy.
Keywords: Personal bankruptcy law; Home equity protection; Stock market participation; Portfolio
allocation
JEL Classification: D14; G00; G11; K35
Resumen: Este trabajo examina empíricamente el efecto del nivel de protección de bancarrota
personal en los Estados Unidos en la demanda de activos financieros de los hogares. Una bancarrota
bajo el Capítulo 7 permite proteger el valor acumulado de la vivienda hasta cierto límite o "exención".
La literatura previa muestra que tal protección sesga la inversión de los hogares hacia la vivienda. Este
trabajo testea si también reduce la inversión en acciones, que no están protegidas en bancarrota. Usando
un enfoque de variables instrumentales, estimo una menor participación en el mercado de acciones
cuando el valor acumulado de la vivienda es menor a la exención, pero el resultado no es robusto, y los
hogares con mayor riesgo de bancarrota no exhiben una respuesta más fuerte. Más aún, la inversión en
vivienda no es mayor cuando la propiedad está completamente protegida. Estos resultados sugieren la
ausencia de distorsiones sustanciales provenientes del nivel de valor de la vivienda protegido en
bancarrota.
Palabras Clave: Ley de bancarrota personal; Protección del valor de la vivienda; Participación en el
mercado de acciones; Asignación del portafolio
*For insightful discussions, I am especially grateful to Christopher Woodruff and Sascha Becker, and also to
Frank Stafford, Francisco Palomino, and Pablo Beker. For helpful advice, I thank Wiji Arulampalam, Dan
Bernhardt, Clément de Chaisemartin, Jorge Luis García Ramírez, Francois Gourio, Hans K. Hvide, Hilary
Hoynes, Peter Hudomiet, Dave Knapp, Gretchen Lay, Alexander Michaelides, Edie Ostapik, Roberto Pancrazzi,
John A. E. Pottow, Mika Vaihekoski, Thijs van Rens, Fabian Waldinger, Jeffrey Wooldridge, anonymous
referees, and participants at several seminars and conferences.
† Dirección General de Estabilidad Financiera. Email: mariela.dalborgo@banxico.org.mx.1. Introduction
The personal bankruptcy law is one of the largest social programs in the US and the most
generous bankruptcy law towards debtors in the world. On average 5 per 1,000 individuals
have started a personal bankruptcy case per year between 1999 and 2011. Its generosity
depends on the amount of resources that consumers can keep after filing for bankruptcy.
Determining the optimal level of protection is a complex issue of great policy interest, given
its potential welfare implications (Livshits et al., 2007; Dávila, 2020). Further, the bankruptcy
law also specifies which resources can and cannot be protected from seizure by creditors.
Such selectivity may have ex ante unintended consequences for the portfolio allocation of
households.
Under Chapter 7, the most common choice to file for bankruptcy, households discharge
most unsecured debt and retain their income. In exchange, they lose their assets, but the
home equity, retirement accounts, vehicles, and bank deposits are protected up to a limit
or exemption. Since homes are generally the main asset in households’ portfolios, the home
equity protection is effectively the largest. Corradin et al. (2016) documents that the so-called
“homestead exemption” biases households’ portfolios towards home equity.
This paper examines empirically a counterpart of that home equity bias. Specifically,
it studies the impact of the homestead exemption in the demand for stocks held outside of
retirement accounts—including those held via mutual funds or investment trusts—that are
not protected in bankruptcy. I conjecture that when the homestead exemption becomes larger
than the home equity, households’ stock holdings will decline in response to a “substitution
effect”: Stocks are lost in bankruptcy and crowd-out investment in protected assets. This
could result in a suboptimal portfolio allocation. Stocks offer higher liquidity than real estate
and diversification gains, thanks to the low covariance of their returns with those of housing
(Jordà et al., 2019).1 Ultimately, addressing this question adds evidence on the effect of
institutional factors on stock market participation. Their role is not well documented yet,
despite that they are one of the explanations from the standard financial theory for the limited
participation puzzle.
A simple portfolio choice model illustrates the proposed mechanism. The household can
invest in home equity or in stocks after paying a fixed cost for participating in the stock mar-
ket. A large negative wealth shock can occur with a positive probability. If the shock occurs,
1 If
the rates of return on stocks are higher than on housing, lower stock holdings may also lower the return
on households’ portfolios. Jordà et al. (2019) find evidence that their returns are similar in the long-run, but
since World War II equities have outperformed housing on average.
1the household defaults on its loans. In exchange, it loses the investment in stocks and any
home equity that is above the homestead exemption. The model predicts that, conditional
on wealth, the household holds more home equity and, hence, less stocks when the home
equity is below the exemption level. In addition, if the exemption level is lower than house-
hold wealth, marginal increases in the exemption reduce stock market participation further.
In both cases, the household response is driven by the purpose of increasing consumption in
bankruptcy, which depends not only on the state exemption level but also on the pre-existing
holdings of home equity.
To test the model predictions, I use longitudinal household data from the Panel Study of
Income Dynamics (PSID) from 1999 to 2011. First, I examine if stock holdings are lower
when the home equity is below the state exemption than when it is above. Given the endo-
geneity of housing wealth and financial assets, ordinary least squares (OLS) estimates may be
biased. Thus, I construct a simulated instrument using nationally representative demographic
groups, adapted from Currie and Gruber (1996) and Mahoney (2015). For every period, I
compute an indicator of whether the average home equity of each group is below or above
the exemption of each state. This removes the variation resulting from idiosyncratic shocks
to households actually living in a given state. The benchmark specification compares house-
holds from the same state and with a similar level of home equity, changing from above to
below the exemption (and vice versa). The resulting variation is mostly driven by changes
in the exemption, which are plausibly exogenous—I verify that they are uncorrelated with
changes in other factors that could potentially drive the demand for stocks. While this strat-
egy removes most confounding factors, the results only provide suggestive evidence since the
potential for omitted variable bias cannot be fully ruled out.
The evidence indicates that stock market participation and the dollar amount invested
in stocks, conditional on participation, are lower when home equity is below than when it
is above the exemption. However, these results are not robust to specifying the model in
first differences in order to remove household-level heterogeneity. Moreover, I do not find
a stronger response, neither at the extensive nor intensive margins, from households more
at risk of bankruptcy (namely, the self-employed, households in bad health, and those with a
middle-aged head). If stock holdings decline because of a substitution effect, the home equity
should increase when the home becomes fully protected. Only the model in first differences
renders a positive and significant effect on home equity, but not the one in levels. When I
replicate the estimates in Corradin et al. (2016) using my simulated instrument, the effect
turns out insignificant.
Next I examine if marginal increases in the exemption affect stock market participation.
2A negative effect is expected only when the exemption is higher than the home equity but
lower than the household’s wealth. This turns out to be a stringent condition, since only few
households have high wealth but low home equity relative to the state exemption. I exploit
the fact that states set different levels of exemptions at different times, which allows dealing
with the possible correlation between state exemptions and unobservable characteristics. Re-
stricting the sample to high-wealth households and allowing for nonlinear marginal effects,
I find a significant decline in participation at intermediate exemptions. For a one-standard
deviation increase in exemptions, participation declines by 6 p.p., which represents a 15%
change relative to the sample mean of the dependent variable (40%). However, no decline is
estimated at higher exemptions and no stronger results are estimated for households at higher
risk of bankruptcy. Not surprisingly, these findings suggest that marginal increases in the
bankruptcy protection do not affect stock market participation.
This paper relates to the literature looking at the effects of insurance provision on house-
hold risk-taking and portfolio choice. Gollier et al. (1997) show theoretically that risk-taking
is higher in the presence of a guaranteed minimum wealth or limited liability. Gormley
et al. (2010) find a positive correlation between access to different types of formal insur-
ance and participation rates. Studies specific to health insurance confirm its positive effect
on stock holdings (Atella et al., 2012; Goldman and Maestas, 2013; Christelis et al., 2020).
Bankruptcy protects against several risks, including that of medical expenses—in fact, it acts
as an informal health insurance, as shown by Mahoney (2015). However, unlike that litera-
ture, this paper does not study whether the bankruptcy protection increases participation via
a “consumption-floor effect”.2 While that mechanism is plausible, it may arise when com-
paring households with and without bankruptcy protection.3 Instead, I study the impact of
the generosity of Chapter 7 in a setting where all households have some bankruptcy pro-
tection. This paper also relates to the household finance literature that studies investment
decisions in the presence of both housing and risky financial assets (Flavin and Yamashita,
2002; Cocco, 2004; Yao and Zhang, 2005; Chen and Stafford, 2016). Cocco (2004) uncov-
ers a “background risk effect” wherein house price risk crowds out stock holdings. Here I
do not examine the impact of real estate risk on the demand for financial risks, but whether
2 Guaranteeing a minimum consumption in bankruptcy decreases exposure to background risk and this
might increase the demand for risky assets (Elmendorf and Kimball, 2000). Background risk refers to sources
of risk different from the volatility of the returns, such as labor and entrepreneurial income volatility, unemploy-
ment and out-of-pocket medical expenses. Even if these risks do not materialize, they can reduce the demand
for risky financial assets ex ante (Kimball, 1993).
3 In European countries, where consumer bankruptcy processes are less common and less generous than in
the US, households not only hold less unsecured debt but also own less stocks than their US counterparts, even
conditioning on characteristics (Christelis et al., 2013, 2017).
3protecting home equity from seizure in bankruptcy has a substitution effect on unprotected
stocks.
This paper builds on the literature showing that the state exemptions impact on the finan-
cial benefit from filing and, ultimately, on households’ bankruptcy decisions. The evidence
on the effect of exemptions on the decision to file for bankruptcy is mixed; some studies find
no effect (Lefgren and McIntyre, 2009) and others a positive effect (Fay et al., 2002; Agarwal
et al., 2003; Lehnert and Maki, 2007; Miller, 2019).4 The result in Miller (2019) that such
positive effect is increasing in households’ assets gives plausibility to the mechanisms pro-
posed here. This paper complements another important strand of the bankruptcy literature,
which studies how it influences entrepreneurship decisions.5 Its most direct contribution is
to the strand considering the effects of the bankruptcy protection on the composition of bor-
rowing (Gropp et al., 1997; Severino and Brown, 2017) and on asset allocation (Greenhalgh-
Stanley and Rohlin, 2013; Corradin et al., 2016). Greenhalgh-Stanley and Rohlin (2013) find
a positive impact of that protection on the housing wealth of older households. However, they
estimate a stronger response when the home equity is high and, therefore, more likely to be
above rather than below the exemption, as predicted here. When the home equity is below
the exemption, Corradin et al. (2016) find a larger investment in home equity in response to
an increase in household wealth. The findings in Corradin et al. (2016) are consistent with
my predictions.
One possible reason for the lack of evidence on stock holdings is that the PSID data have
little statistical power to capture that effect. However, it is striking that I do not even find a
significant effect on home equity, where statistical power is less of a concern according to the
test proposed by Burlig et al. (2020). If power is low even for home equity (Corradin et al.,
2016, use data from the Survey of Income and Program Participation, or SIPP, which samples
a larger number of households), my results suggest that the home equity bias is not of first or-
der to show up in a representative survey of US households. On the other hand, my empirical
strategy enhances the estimates by Corradin et al. (2016) in several dimensions, challenging
the existence of such bias. First, they instrument the indicator for home equity being below
the exemption with the exemption level in 1920. I instead build a time-varying instrument
from nationally representative groups that deals with the main source of endogeneity coming
4 In the theoretical literature, Athreya (2006), Pavan (2008), Lopes (2008), Mankart (2014), and Mankart
and Rodano (2015) predict a positive relationship, whereas Li and Sarte (2006) and Mitman (2016) predict a
negative one.
5 See, e.g., Berkowitz and White (2004); Berger et al. (2011); Fossen (2014); Mankart and Rodano (2015);
Cerqueiro and Penas (2016).
4from changes in home equity.
Second, they provide evidence of heterogeneous effects by showing a slightly larger re-
sponse from households in bad health and younger. I test for significant differences across
groups, which provides more convincing evidence, and look at middle-aged households,
which are the ones more at risk of bankruptcy according to Fisher (2019). Third, the authors
study whether investment in housing is differently affected by changes in household wealth
when the home equity is below the exemption. While intuitively plausible, this adds an ad-
ditional source of endogeneity.6 I investigate the pure effect of having home equity below
the exemption, conditioning on wealth. This mitigates the possibility of estimating a spuri-
ous positive relationship, for instance, if the elasticity of home equity investment to wealth
changes is larger when the home equity is low. In summary, these considerations, along with
my findings, cast doubts on a home equity bias attributed to the bankruptcy protection. The
absence of such bias in the first place can account for why the bankruptcy protection has no
effect on stock holdings.
2. US Personal Bankruptcy Law
Individuals smooth consumption over the life-cycle by taking loans in the presence of un-
certainty. If income turns out to be low or expenses high, individuals would have to reduce
consumption dramatically or will not be able to meet their financial obligations. In that con-
text, the bankruptcy law has two main conflicting functions. One is to act as a consumption
insurance, by allowing debtors to discharge most unsecured debt, including credit card debt,
installment loans, and medical bills. Debts that are not dischargeable include tax obligations,
student loans, alimony, child support obligations, debts incurred by fraud, credit card debt
incurred just before filing, and some secured debt such as mortgages and car loans.7 The
second function of the bankruptcy law is to discourage households from borrowing without
considering if they are or will remain solvent. This is achieved by imposing costs for de-
faulting, which include future exclusion from credit markets, the ban to file again for several
years, and filings becoming public knowledge and appearing on credit records for ten years.
In exchange for discharging debt, there are two repayment options. Under Chapter 13, house-
6 Inturn, wealth changes can affect the probability of having home equity below the exemption (both are
measured at period t). Thus, adding their interaction gives rise to the “bad control” problem (Angrist and
Pischke, 2009).
7 Even after filing for Chapter 7 the borrower has to continue making mortgage payments, otherwise the
lender can foreclose the house. However, Chapter 7 prevents the lender from going after the borrower’s personal
assets when the house is sold through foreclosure and the sale proceeds are lower than the mortgage balance.
5Table 1: Average exemption levels for couples and singles, 1999-2011
Average exemption (excl. states States with unlimited
with unlimited exemption) ($) exemption Average home equity ($)
Couples Singles Couples Singles
1999 77,163 50,324 AR, FL, IA, KS, OK, SD, TX 55,496 19,636
(75,835) (63,950) (31,182) (17,609)
2001 71,226 47,022 AR, DC, FL, IA, KS, OK, SD, TX 65,492 28,707
(66,021) (55,491) (38,353) (38,662)
2003 68,654 48,535 AR, DC, FL, IA, KS, OK, SD, TX 74,081 33,462
(69,725) (66,676) (54,306) (33,402)
2005 69,736 49,486 AR, DC, FL, IA, KS, OK, SD, TX 78,886 30,858
6
(85,598) (80,526) (54,732) (22,370)
2007 76,588 53,449 AR, DC, FL, IA, KS, OK, SD, TX 82,230 32,278
(96,096) (90,435) (56,094) (23,628)
2009 112,992 92,429 AR, DC, FL, IA, KS, OK, SD, TX 64,878 28,716
(157,834) (159,664) (38,236) (41,354)
2011 134,152 109,009 AR, DC, FL, IA, KS, OK, SD, TX 61,677 20,050
(180,337) (185,462) (40,312) (21,870)
1999-2011 87,183 64,943 68,963 27,634
(114,030) (113,642) (46,199) (29,931)
Notes. This table shows the average homestead plus wildcard exemption levels for couples and singles in real 2004 dollars (deflated using the
state house price index) for states with no unlimited homestead exemption. States with unlimited homestead exemption by year are listed in
the middle column. The last two columns report the average home equity for couples and singles in real 2004 dollars. Standard deviations are
reported in parenthesis. Data are extracted from bankruptcy filing manual books (Renauer et al., several editions).holds pay out of post-bankruptcy income over the following three to five years but can keep
all their assets. Under Chapter 7, debtors lose their assets but can keep future income. To
provide extra insurance to households, Chapter 7 sets exemptions for some asset categories.
Goodman (1993) explains that the asset exemptions emerged during the second half of the
nineteenth century to attract population into uninhabited areas. Regions more interested in
attracting potential settlers set higher exemptions. The exempt assets usually comprise equity
in owner-occupied homes, vehicles, retirement assets, and bank deposits. Some states also
have “wildcard exemptions” that can be used to protect non-exempt assets or assets in excess
of the corresponding exemption. The exemptions vary by state and marital status, and on
occasions also by age (65 or over) and disability status. Some states give the option to file
under the exemptions set by the federal law.
During the period from 1999 to 2011, the Chapter 7 exemptions have experienced sub-
stantial variation over time. Table 1 shows that the average (homestead plus wildcard) ex-
emption for couples increased from $77,163 in 1999 to over $134,000 in 2011 (in real 2004
dollars). Washington DC was the only state that changed to an unlimited homestead exemp-
tion over those years, starting from the federal level. Differences in exemption levels across
states are also very large, as reflected in the large standard deviations. For each state, Figure
1 shows the logarithm of those exemptions in 1999 and 2011. Since some states, such as
Vermont, Wyoming, Virginia, exhibit little or no variation in nominal values, they experience
a decline in real terms over that period. In others, such as Oregon, Louisiana and Maine,
the increase in nominal terms was just enough to account for the increase in home prices. In
contrast, the increase of those at the bottom of the figure, such as Nevada, Rhode Island and
South Carolina, has been substantial.
Under Chapter 7, unsecured debt is discharged by the end of the case. “Asset cases”
are those in which the filer owns non-exempt assets (i.e. assets above the corresponding
exemption or from non-exempt categories), which become property of the bankruptcy estate
by the time of filing. A trustee assigns a dollar amount to the non-exempt assets and divides
the funds among creditors. In the more common “non-asset cases”, the filer does not own
non-exempt assets. Asset cases usually close about two years after filing and non-asset cases
close within four months. The practice of selling any non-exempt asset before filing, known
as “bankruptcy planning”, is discouraged by the law when its purpose is to defraud creditors.
Proceeds can still be used to pay the attorney and other filing fees. Table 1 shows that the
average home equity level was below the average exemption for both couples and singles
between 1999 and 2011—only in 2003, 2005 and 2007 it was above the average exemption
for couples.
7Figure 1: Logarithm of state exemption levels for couples, 1999-2011 (2004 USD)
1999 2011
Vermont
Wyoming
Virginia
Connecticut
Hawaii
Alaska
North Dakota
New Jersey
Kentuky
Mississippi
Alabama
California
Pennsylvania
Oregon
Lousiana
Maine
West Virginia
Michigan
Montana
Arizona
Minnesota
New Mexico
Tennessee
Missouri
Colorado
Utah
Illinois
Idaho
New Hampshire
Georgia
Indiana
North Carolina
Washington
Wisconsin
Massachusetts
Maryland
Nebraska
Ohio
New York
Nevada
Rhode Island
South Carolina
8 9 10 11 12 13 14
Log homestead + wildcard exemption
Notes. This figure shows the logarithm of the sum of the homestead plus wildcard exemptions in
1999 and 2001 in real 2004 dollars (deflated using the state house price index). It excludes states with
unlimited homestead exemptions (Arkansas, Florida, Iowa, Kansas, Oklahoma, South Dakota, Texas
and District of Columbia) and Delaware that was an outlier for the first years of the sample. Data are
extracted from bankruptcy filing manual books (Renauer et al., several editions).
8Before 2005 debtors could choose under which chapter they wanted to file, and the most
common choice was Chapter 7. Thus, even defaulters with very high incomes were not
committed to future repayments. Under Chapter 13 borrowers repay from post-bankruptcy
income, a less attractive alternative. This system encouraged strategic behavior and became
beneficial for individuals with high income and wealth. In addition, Chapter 13 filers were
able to propose their own repayment plans and typically proposed an amount equal to the
value of their non-exempt assets: They were not allowed to repay less and since they had the
option to choose Chapter 7, they had no incentives to repay more (White, 2007). This means
that even for those who decided to file under Chapter 13, Chapter 7 exemptions would still
affect the repayment amount and therefore the probability of declaring bankruptcy.
In 2005 there was a reform in the law, known as the “Bankruptcy Abuse Prevention and
Consumer Protection Act of 2005” (BAPCPA). Its purpose was to restrict the speculative
behavior that led to historically high levels of bankruptcy. The reform removed the debtor’s
right to choose between Chapters 7 and 13. In order to qualify for Chapter 7, debtors need to
pass a “means test”, except for the self-employed with mostly business debt and for certain
members of the military. To pass that test, the annualized average income over the six months
before filing needs to be smaller than the state median income. Otherwise, the “disposable
income” needs to be smaller than a certain amount.8 Thus, high-income households have
to file under Chapter 13 and can no longer propose their own repayment plans under that
chapter. The reform also increased the costs of filing under both chapters and extended the
minimum time that debtors must wait before filing again. In addition, it introduced a new
exemption for tax-protected individual retirement accounts (up to $2 million for couples and
$1 million for singles).
Figure 2 shows that the number of Chapter 7 filings was on average 3.6 per 1,000 in-
habitants between 1997 and 2004. That figure increased by more than 50% to 5.5 in 2005,
when debtors who anticipated going bankrupt had an incentive to file before the new law was
implemented. After the reform, the number of filings plummet to just 1.2 in 2006 and then
started to rise, reaching the pre-reform levels by 2008.
8 “Disposable income” is the difference between debtors’ average monthly family income during the six
months prior to filing and a new income exemption, which determines an allowance for living expenses.
9Figure 2: Number of Chapter 7 filings per capita by year
6
5
4
3
2
1
0
Notes. The columns show the number of non-business Chapter 7 filings per 1,000 inhabitants (an-
nual averages) extracted from the Statistics Division of the Administrative Office of the United States
Courts.
3. Theoretical Framework and Empirical Predictions
3.1. A model of portfolio choice with bankruptcy
To derive the empirical predictions, this section presents a two-period model of household
portfolio choice in the presence of a bankruptcy protection similar to that from Chapter 7. The
household starts the first period with initial wealth W that can invest in two asset categories,
home equity (h) and a risky financial asset (s). If it invests in the risky asset, it faces a fixed
cost of stock market participation, q.9 The net return on h equals 0 and on s is random and
equals rhigh > 0 with probability p > 0 and −rlow < 0 with probability 1 − p.10 I define the
stock premium as the difference in the expected net returns of s minus h and assume that it is
positive to induce stock market participation, that is, prhigh − (1 − p) rlow > 0.
In the second period, the household faces a large negative wealth shock and declares
9 Fixed costs of entry and participation into the stock market are assumed, for example, by Vissing-Jorgensen
(2002), Haliassos and Michaelides (2003), Gomes and Michaelides (2005) and Alan (2006).
10 Even though housing is also a risky asset, it can be assumed as safer than equities. Jordà et al. (2019) show
that its return has a lower volatility and lower covariance with consumption growth than the return on equities.
Corradin et al. (2016) assume that non-housing assets have higher returns than housing and that both are safe
assets.
10bankruptcy with probability ε. Further, I assume that ε is uncorrelated with p (the shock can
occur, for instance, if an unexpected illness leads to a substantial increase in medical expenses
or if the individual becomes unemployed and experiences a large loss of future income). If the
shock occurs, the household does not have to repay its loans and second-period expenses are
zero rather than positive. In exchange, it loses all the investment in the risky asset, whereas
it can guarantee a minimum level of consumption given by h, up to a maximum of H > 0.
With probability 1 − ε, the household does not face that major shock and, therefore, does not
declare bankruptcy.
The optimal level of stocks denoted by s∗ includes the possibility of nonparticipation
(i.e., s∗ = 0). If s∗ = 0, the household invests all its wealth into home equity, h∗ = W , and
consumption in the second period equals W with probability 1 − ε and min W, H with prob-
ability ε. Thus, when the bankruptcy exemption H is low, the household consumes its initial
wealth in the good state and H in the bad state. When that exemption is high, it consumes its
initial wealth in both states.
If s∗ > 0 and assuming that the utility function is of logarithmic form, U (C) = log(C), in
the first period the household chooses h and s to solve the problem given by:
maxE [log (C)] (1)
h,s
subject to the following constraints:
h+s+q ≤W (2)
h + (1 + rhigh )s
1 − ε, p
C= h + (1 − rlow )s 1 − ε, 1 − p (3)
min h, H
ε
C≥0 (4)
h≥0 (5)
s≥0 (6)
where I assume that the time discount rate equals 1 without loss of generality. The budget
constraint in equation (2) will hold with equality, given the assumption of nonsatiation. Con-
dition (3) gives the three possible consumption levels with their corresponding probabilities
in the second column. The inequality in (4) is never binding since limC→0 C1 = ∞. The con-
11straints in (5) and (6) reflect the assumption that the quantities of the two assets held are
nonnegative. They ensure that the share of wealth invested in the risky asset is bounded be-
tween 0 and 1. If the constraint (5) were binding, the condition (3) implies that C = 0 with
probability ε. Thus, home equity at a corner of zero cannot be optimal since limC→0 C1 = ∞.
In turn, since the household participates in the stock market, the condition (6) is not bind-
ing. The solution to the problem given by equations (1) to (6) is denoted as s∗∗ , which is the
optimal level of stock holdings conditional on participation.
I denote by V in and V out the indirect utilities of households that participate and do not
participate in the stock market, respectively. The household participates when it gives higher
utility than investing purely in home equity:
V = max V in ,V out
(7)
1(s∗ >0)
where 1(s∗ > 0) is an indicator function that equals 1 if the household participates in the stock
market and 0 otherwise. The indirect utility of not participating equals V out = (1 − ε) log (W ) +
+ε log min W, H . Under these conditions, I obtain the following proposition:11
Proposition 1. For a given level of wealth, the probability that the household participates
in the stock market is smaller when the home equity, conditional on participation, is fully
protected in bankruptcy than when it is not. This effect is increasing in the probability of
bankruptcy.
The household never participates in the stock market for high q and always participates
for low q. For the same wealth level, an increase in H that moves the solution from the
region where h∗∗ ≥ H to that where h∗∗ < H reduces the threshold for q that determines
participation. This leads to Proposition 1, which implies that participation is lower when the
exemption level becomes high relative to the home equity, conditional on participation. The
lower participation when h∗∗ < H occurs because holding s reduces h and, therefore, reduces
consumption in bankruptcy. This effect is larger when the probability of bankruptcy is higher.
Marginal changes in the bankruptcy exemption can also affect the participation decision, as
stated by the following proposition:
Proposition 2. Marginal increases in the exemption reduce the probability of participa-
tion when the exemption is larger than the optimal home equity, conditional on participation,
11 All proofs can be found in Appendix B.
12but smaller than the wealth level, h∗∗ < H < W . This effect is stronger when the probability
of bankruptcy is larger.
Proposition 2 allows identifying three regions. When h∗∗ ≥ H, increases in the exemption
do not affect the probability of participation; they lead to higher consumption in bankruptcy
irrespective of whether the household holds stocks. An increase in H that leaves the home
fully protected, h∗∗ < H, reduces discretely the probability of participation, as stated by
Proposition 1. Proposition 2 predicts that it continues declining as the exemption increases
further; in that region, holding stocks prevents increasing consumption in bankruptcy as the
exemption becomes larger. The higher the probability of a negative shock, ε, the more de-
creases the probability of participation. At high exemptions, when H > W , participation is
no longer affected by marginal increases in the exemption; the household has already in-
vested all its wealth in the protected asset. Finally, the third proposition describes the optimal
investment decision at the intensive margin in the presence of bankruptcy protection:
Proposition 3. For a given level of wealth, if the stock premium is smaller than rhigh ×
rlow , the optimal value invested in stocks (s∗∗ ), conditional on participation, is smaller when
the home equity is fully protected in bankruptcy (h∗∗ < H) than when it is not. In addition,
when h∗∗ < H, s∗∗ is decreasing in the probability of bankruptcy.
Proposition 3 states that, when the expected return of the financial asset is low, an increase
in H that moves the solution from the region where h∗∗ ≥ H to that where h∗∗ < H increases
the attractiveness of investing in home equity and, therefore, reduces investment in stocks at
the intensive margin. If the expected return of the financial asset is high, investment in that
asset will always be high and no decline is expected following an increase in H. The second
part of Proposition 3 says that, when h∗∗ < H, a higher probability of a negative shock reduces
the amount invested in the financial asset, since it reduces its marginal benefit and increases
its marginal cost. When h∗∗ ≥ H, that effect is not expected because the household cannot
increase consumption in bankruptcy by investing more in home equity.
I illustrate the former predictions with an example, using a standard parameterization.
I consider the case where rhigh =1 and rlow = 1, which corresponds to net returns (1, −1),
similar to the sum of two “Arrow-Debreu” assets. The resulting stock premium, 2p − 1, is
positive whenever p > 0.5, so I set p = 0.75. I assume that the wealth level equals the average
wealth in the sample, W = 200 (in thousands), and that the participation cost equals q = 10.
Table 2 shows the results at different exemption levels, for ε = 0.05 in the middle section and
for ε = 0.15 in the bottom section.
13The middle section shows that qI , the threshold for q above which the household does not
participate in the stock market, does not change when the exemption goes from H = 50 to
H = 80 and the home is not fully protected (h∗∗ = 95). When the home becomes fully protec-
Table 2: Numerical example
H = 50 H = 80 H = 150 H = 180 H = 200 H = 250
rhigh 1 1 1 1 1 1
rlow 1 1 1 1 1 1
p 0.75 0.75 0.75 0.75 0.75 0.75
W 200.0 200.0 200.0 200.0 200.0 200.0
q 10.0 10.0 10.0 10.0 10.0 10.0
ε 0.05 0.05 0.05 0.05 0.05 0.05
hI 87.7 87.7 103.3 77.0 77.4 77.4
sI 87.7 87.7 76.3 104.2 104.8 104.8
qI 24.5 24.5 20.4 18.8 17.8 17.8
h∗∗ 95.0 95.0 109.3 109.3 109.3 109.3
s∗∗ 95.0 95.0 80.8 80.8 80.8 80.8
ε 0.15 0.15 0.15 0.15 0.15 0.15
hI 87.7 87.7 133.7 137.4 139.6 139.6
sI 87.7 87.7 50.7 52.1 52.9 52.9
qI 24.5 24.5 15.6 10.5 7.5 7.5
h∗∗ 95.0 95.0 137.8 137.8 200.0 200.0
s∗∗ 95.0 95.0 52.3 52.3 0.0 0.0
Notes. This table shows a numerical example for the predictions of the theoretical model in section
3.1. Each column corresponds to a different exemption level, H. rhigh and −rlow denote the net rates
of return on the risky asset, p is the probability of a high return and W is the wealth level. q is
the participation cost so that the household participates if q < qI and does not participate otherwise.
ε is the probability of a negative wealth shock. hI and sI are the levels of home equity and stock
holdings corresponding to a participation cost of qI that equalizes the expected utility of participating
and non-participating. h∗∗ and s∗∗ are the optimal investment in home equity and in stocks conditional
on participation for a given cost q. All dollar values are expressed in thousands. The center panel
presents the results assuming ε = 0.05 and the bottom panel the results for ε = 0.15. The parameters
in the top panel are used for both sets of results.
14ted at H = 150, qI declines, which implies a lower participation as predicted by Proposition
1. qI keeps declining as H increases to 180 and then to 200, as predicted by Proposition 2. It
remains the same at exemption levels beyond H = 200, that is, beyond the wealth level, W .
In turn, the optimal level of stocks conditional on participation declines from s∗∗ = 95 when
the home is not fully protected to s∗∗ = 81 when it becomes fully protected following an
increase in H, as predicted by Proposition 3. Finally, the bottom section presents the results
for ε = 0.15. When the exemption is low, H = {50, 80}, the results are the same as when
ε = 0.05. But the declines in qI and s∗∗ after the home equity becomes fully protected are
more pronounced, consistent with the predicted higher sensitivity of households more at risk
of bankruptcy at both the extensive and intensive margins.
3.2. Empirical predictions
Propositions 1 to 3 predict that a higher homestead exemption under Chapter 7 has a non-
linear, negative effect on the demand for stocks outside retirement accounts, which are lost in
bankruptcy. The non-linearity comes from the fact that the effect of the bankruptcy exemption
depends on the household’s wealth level and on how much of that wealth is invested in home
equity.12
On the basis of the propositions of section 3.1, I state the main hypotheses that will guide
the empirical tests in section 5.
- Hypothesis 1. Conditional on household’s wealth, the probability of participating in the
stock market is smaller when the home equity is fully protected in bankruptcy than when it is
not. That differential is increasing in the probability of bankruptcy.
- Hypothesis 2. The probability of participating in the stock market is decreasing in the
exemption level when that level is higher than the home equity but lower than the household’s
wealth. This effect is stronger when the probability of bankruptcy is larger.
- Hypothesis 3. Conditional on household’s wealth and on stock market participation,
investment in stocks is smaller when the home equity is fully protected in bankruptcy than
when it is not. That investment is decreasing in the probability of bankruptcy when the home
12 Propositions 1 to 3 predict no positive income effect on the demand for stocks from the bankruptcy pro-
tection. However, it can be shown the presence of a positive consumption floor effect on participation relative
to a setting without such protection. In the absence of bankruptcy protection, the household has to repay a large
negative wealth shock, −D < 0, in the second period. As a result, the third row of the budget constraint in (3)
changes to C = h + (1 + rhigh )s − D with probabilities ε and p and to C = h + (1 + rhigh )s − D with probabilities
ε and 1 − p. The value of q that leaves the household indifferent between participating or not in the stock
market is smaller than the values derived in Proposition 1. Thus, the region where non-participation is optimal
is larger in the absence of bankruptcy protection. This result, however, cannot be tested empirically exploiting
cross-state variation in the US, given the ubiquity of the Chapter 7 protection.
15equity is fully protected.
4. Data and Sample Definition
The source of household data is the PSID, a longitudinal panel survey that is representative
of the entire population of the United States and their families. It has detailed information on
portfolio composition, publicly available state identifiers, and household socio-demographic
characteristics.13 Asset holdings are from the wealth survey, conducted every other year
since 1999. Thus, I define the sample for the period between 1999 and 2011. It is restricted
to households where the head is 65 years old or younger in every year—households hold more
conservative portfolios as they age and are less likely to file for bankruptcy. After excluding
observations where some of the regressor variables are missing, the final sample has 46,454
observations corresponding to 50 states plus Washington, D.C.
Table 3 reports descriptive statistics for all household and state variables (see Appendix
Table A.1 for definitions). The statistics are reported for the entire sample and, separately,
for households with home equity below and above the (homestead plus wildcard) exemption
level in a given year. This sample split means that households can change their classification
over the period. About 70% of the observations have home equity below the exemption.
Table 3: Summary statistics, 1999-2011
All households Below exemption Above exemption
N = 46,454 N = 33,046 N = 13,408
Panel A: Household-level data
Stock market participation 0.17 0.11 0.31
(.37) (.31) (.46)
log(Stocks) ($) 9.87 9.25 10.39
(2.14) (2.19) (1.94)
IHS(Home equity) ($) 10.37 8.53 12.09
(4.82) (6.38) (.90)
IHS(Wealth) ($) 7.40 5.32 12.51
(8.05) (8.55) (2.71)
13 The PSID offers more tractability for panel data analysis than the SIPP, which is composed of a series of
multiyear panels and, hence, does not track the same households through an extended period.
16All households Below exemption Above exemption
N = 46,454 N = 33,046 N = 13,408
IHS(Income) ($) 9.70 9.44 10.33
(3.63) (3.66) (3.46)
log(Age of the head) (years) 3.66 3.60 3.82
(.31) (.31) (.24)
College education 0.49 0.45 0.59
(.50) (.50) (.49)
Highest year of college 1.48 1.28 1.97
completed (1.81) (1.71) (1.95)
Married 0.59 0.52 0.75
(.49) (.50) (.43)
log(Family size) 0.88 0.85 0.98
(.57) (.59) (.50)
Self-employed 0.10 0.08 0.13
(.30) (.28) (.33)
Bad health 0.14 0.15 0.10
(.34) (.36) (.30)
Age 35-49 years old 0.39 0.37 0.43
(.49) (.48) (.5)
Panel B: State-level data
log(Exemption) ($) 11.31 11.75 10.22
(excl. unlimited) (2.20) (2.41) (.86)
Unlimited exemption 0.16 0.23 0.00
(.37) (.42) -
log(Inflation-adjusted -0.10 -0.11 -0.08
house price) (.17) (.17) -
log(Unemployment rate) -2.82 -2.80 -2.86
(.36) (.37) (.34)
log(Proprietor employment) 13.68 13.70 13.65
(.85) (.9) (.71)
log(Per capita personal 10.42 10.41 10.43
income) ($) (.14) (.15) (.14)
17All households Below exemption Above exemption
N = 46,454 N = 33,046 N = 13,408
log(State GDP) ($) 12.79 12.78 12.81
(.89) (.93) (.77)
log(Nr. of non-business 1.49 1.47 1.54
bankruptcy filings/1,000 inh.) (.46) (.46) (.44)
log(Per capita medical 8.56 8.56 8.56
expenses) ($) (.16) (.17) (.15)
Notes. Household-level data correspond to household heads in the 1999 to 2011 PSID panels, 65 years
old or younger. The first column includes all the sample and the next two columns split the sample in
households with home equity below and above the (homestead plus wildcard) state exemption in year
t. Monetary values are in real 2004 dollars and winsorized at the 1st and 99th percentile. All variables
are described in Appendix Table A.1. Standard errors are in parentheses.
Panel A shows households’ asset holdings and socio-demographic characteristics. Only
11% of households with home equity below the exemption hold stocks outside retirement ac-
counts on average, versus 31% of those with home equity above the exemption. The measure
of stocks includes those invested in publicly held corporations, mutual funds, or investment
trusts, and excludes those invested into individual retirement accounts. Retirement accounts
can be subject to deposit limits and to penalties from early withdrawal and are protected from
bankruptcy up to $1 million since 2005 (subject to cost-of-living adjustments). Conditional
on ownership, the amount held in stocks is also significantly lower for households below the
exemption. One caveat is that the dollar amount held in stocks generally suffers from high
measurement and reporting errors in survey data (Fagereng et al., 2017). In addition, at the
intensive margin, passive variations in the value of stocks may conceal active rebalancing
decisions made by households (Calvet et al., 2009). These issues apply to survey data in
general and, hence, are not unique to the PSID.
Panel A also shows that households on average have lower wealth and income when their
home equity is below than above the exemption. I take the inverse hyperbolic sine (IHS)
transformation of home equity, wealth and income to preserve observations with negative or
zero values (not possible with a logarithmic transformation), as recommended by Burbidge
et al. (1988). In addition, households with home equity lower than the exemption are younger,
less likely to have some college education, and have less years of college on average than
those with home equity above the exemption. They are less likely to be married and live
in households with fewer family members. Finally, they are less likely to be self-employed
18and to have a middle-aged head, and more likely to be in bad health. Being self-employed,
middle-aged, and in bad health are positively correlated with a higher likelihood of filing for
bankruptcy.
Panel B of Table 3 shows summary statistics for the state-level data. The state exemptions
were extracted from bankruptcy filing manual books (Renauer et al., several editions). I
only include homestead plus wildcard exemptions, which represent on average about 90% of
the total asset exemptions (i.e., after adding exemptions for vehicles, business wealth, and
other real estate). The summary statistics for the levels and changes in exemptions are only
computed for states with no unlimited homestead exemption. For the regressions, whenever
the homestead exemption is unlimited, I set the corresponding value to the maximum level
of home equity ever observed in the sample. The third row of Panel B shows that on average
16% of the entire sample and 25% of the households in the below-the-exemption group live
in states with unlimited protection. By definition, there is no household in the above-the-
exemption group living in those states.
The state level variables used as controls are taken from Freddie Mac (the house price
index), the Bureau of Labor Statistics (BLS) (unemployment rate), the Bureau of Economic
Analysis (BEA) (proprietor employment, per capita personal income, real GDP), the Statis-
tics Division of the Administrative Office of the US Courts (non-business bankruptcy filings),
and the Centers for Medicare & Medicaid Services (CMS) (per capita medical expenses).
Data on medical expenses are available only until 2009. I deflate all nominal values by the
BLS Consumer Price Index (CPI) (2004 = 100), using the index for urban consumers as a
proxy for the state-level index.
5. Empirical Analysis
5.1. Stock market participation after the home equity becomes fully protected
5.1.1 Empirical strategy
In this section I test for the first hypothesis that households are less likely to participate in the
stock market when their home equity becomes fully protected under Chapter 7. I estimate the
following model:
Sist = α0 + α1 Belowist−1 + α2 Xist + α3 Rst + αt + αs,t + αs × αh + νist (8)
19where Sist is a dummy indicating whether household i, living in state s in year t owns stocks.
∗ > 0 and S = 0 if S∗ = 0, where S∗ is a continuous latent
That is, we observe Sist = 1 if Sist ist ist ist
variable. Belowist−1 is a dummy that takes the value of 1 if household i, living in state s in
year t − 1, has home equity below the exemption of that state and year and of 0 otherwise.
Under Hypothesis 1, α1 should be negative if participation declines when the home equity
is below the exemption. In the baseline specification, the coefficients are estimated using an
OLS model.14 The error term, νist , absorbs the idiosyncratic variation in stock ownership
across households, states, and time. Since the effect of being below the exemption is likely
to be correlated within a state, I cluster the standard errors at the state level.
I follow the literature by including time-varying control variables with a demonstrated
effect on participation (see, e.g., Guiso et al., 2008; Giannetti and Wang, 2016). The set of
socio-economic and demographic controls, Xist , includes restricted cubic splines of the IHS
transformation of household wealth and labor income to capture nonlinearities in the wealth
and income effects, dummies for whether the head is married and for whether has some
college education, and the logarithm of the head’s age and family size. I control for state-
level variables in Rst to reflect economic conditions, including proprietor employment and
state house prices deflated by the CPI. The latter accounts for the fact that the conditions in
the housing market could be correlated with the bankruptcy protection and drive participation
via wealth effects (Chen and Stafford, 2016).
The constant α0 measures aggregate financial parameters (such as the risky asset pre-
mium). The model also includes a time dummy to control for factors that affect the entire
cross-section of individuals in any given year (αt ). This responds to a common identifying
assumption in this literature that there are age and time effects in participation but no cohort
effects. The baseline model controls for all state-specific factors that are constant over time
and can affect outcomes via state fixed effects (αs ). This accounts for the possible correlation
between the homestead exemption laws and unobserved state-level factors. Additionally, I
account for differential state-specific linear time trends that capture unobserved state charac-
teristics changing over time, αs,t . In the benchmark specification, I saturate the regressions
with pairwise interacted fixed effects between state and household home equity, αs × αh .15
This allows comparing households from a given state with a similar level of home equity that
14 I estimate an OLS model despite that the dependent variable is binary. The large number of fixed effects
over several dimensions would give rise to an incidental parameter problem when using a nonlinear model,
whereas the OLS coefficients are still consistent.
15 α is defined for one-unit bins of the IHS of home equity, which singles out households with negative
h
home equity and renters, rendering a maximum of 14 bins.
20are below and above the exemption.16
Being below the state exemption is endogenous to stock market participation. In par-
ticular, omitted variables could drive both home equity and stock holdings. A household
that experiences a negative wealth shock may reduce its home equity, by taking a home eq-
uity loan for instance, switching from being above to being below the state exemption. If it
also exits from the stock market, α̂1 will be negative for reasons unrelated to the bankruptcy
protection. To address this possibility, I follow the instrumental variable (IV) strategy from
Mahoney (2015), based on Currie and Gruber (1996). The instrument measures the generos-
ity of the homestead exemption in a state and year towards different demographic groups.
By controlling for state and demographic group fixed effects, this so-called “simulated in-
strument” exploits within-state variation in the protection given to the home equity of each
group. Demographic groups that invest a higher fraction of their wealth in home equity are
more likely to be below the exemption in states with a higher homestead exemption. Since the
demographic groups are nationally representative, the variation resulting from idiosyncratic
shocks to households of a given state is removed.
To construct the instrument, I divide the sample into groups on the basis of all possi-
ble combinations of predefined categories for age, race, education, and family structure.17 I
compute the average home equity across all states for each demographic group in every year,
excluding the own value of household i. Then I create a dummy that takes the value of 1 if
the exemption level is above the group home equity for state s in year t, and of 0 otherwise.
A simpler strategy would have been to only use the exemption level as an instrument. Con-
sidering that the IV estimator renders the treatment effect for the compliers, an advantage of
using the simulated instrument is that the compliers are more representative of the average
person in the population.18 For example, a 16-35 years old, minority household, with no
college education, and married with children is likely to be below the exemption generally.
When the exemption is low, such household will be a complier to the simulated instrument,
but not to the exemption level instrument.
16 I do not include individual-level fixed effects because only 21% of the households in the sample change
status from below to above the exemption or vice versa and, moreover, the response in terms of stock holdings
may not be immediate.
17 Similar to Mahoney (2015), the categories for age are 16-35, 36-45, 46-55 and 56-65 years old; for race
are white and non-white; for educational attainment are less than college, some college and college completed;
and for family structure are single and childless, married and childless, single with children, and married with
children.
18 In the simulated instrument, the compliers are households below the exemption if the home equity of their
demographic group is below the exemption and above the exemption if the group home equity is above. Using
the exemption level as the instrument, the compliers will be households below the exemption if the exemption
is high, and above the exemption if it is low.
21Since the endogenous explanatory variable is binary, it is possible to use the two-step in-
strumental variable (IV) method (see Wooldridge, 2010). Thus, I first estimate a probit model
for the Belowist−1 dummy on the simulated instrument, the time-varying controls (Xist and
Rst ), and state and year fixed effects. Households living in states with unlimited exemptions
are dropped because they are perfect predictors of being below the exemption. Then I esti-
mate an ordinary two-stage IV regression of equation (8) using the predicted values of the
probit model as an instrument for Belowist−1 . This approach improves the efficiency and pre-
cision of the two-stage least square (2SLS) estimator and does not require the probit model
to be correctly specified for consistency.
5.1.2 Results
Table 4 shows the OLS estimates of equation (8) and the IV estimates using the simulated
instrument. I restrict the sample to that used for the IV models that drops states with un-
limited homestead exemption.19 When controlling for state and year fixed effects and for
state-specific linear time trends, the OLS model shows that participation is not significantly
different when the home equity is below the exemption than when it is above (column 1).
The 2SLS result is presented in column 2, where the exclusion restriction is given by the fit-
ted values of the first-stage probit for the Belowist−1 dummy on the simulated instrument, all
the observable controls, state and year fixed effects, and state time trends. In the (unreported)
first-stage estimate, the coefficient on the simulated instrument equals 0.326 and is significant
at the 1% level. On the basis of the large F-statistic in column 2, the predicted probability
of being below the exemption is strongly related to the actual probability. The second stage
result shows a negative coefficient on the Belowist−1 dummy (significant at the 1% level),
which suggests that omitted variables bias the OLS estimate upward. This bias could re-
flect unobserved household factors, such as risk aversion or the preference for financial over
real wealth, that are negatively correlated with the decision to accumulate home equity and
positively correlated with the decision to participate in the stock market.
The remaining columns add state×home equity fixed effects to identify more precisely
the effect of having the home fully protected—they control more comprehensively for unob-
served, time-invariant drivers of participation at the state and home equity levels. In column
3, the OLS coefficient becomes negative and statistically significant at the 1% level. The IV
estimate in column 4 becomes only slightly larger than in column 2. Finally, columns 5 and
19 The OLS results remain similar when estimated in the full sample.
22You can also read
